TrustyGenius

Five specialist agents.
One platform.

Each Genius is a focused autonomous analyst — not a chatbot with broker access. They run on schedule, log every insight, and learn what works on your book.

The five specialists

What each Genius does

Alpha Discovery
Hypothesis generation across equities.

Scans equities across three sub-books — momentum, mean-reversion, and cross-sectional anomalies — every morning. Proposes ranked strategy candidates with backtests attached. Each candidate is logged to the Insight Journal and re-evaluated as the regime shifts.

Commodity Cycle Strategist
Spreads and term structure across futures.

Reads inventories, weather, shipping, and contract roll dynamics across energy, metals, and ags. Flags backwardation/contango shifts, congestion-driven dislocations, and seasonal opportunities. Built around the rolling-contract graph in tg_graph.

Macro & Risk Monitor
Regime classification and portfolio guardrails.

Daily regime classification — volatility, trend, correlation, liquidity. Watches your portfolio's exposure, drawdown trajectory, and correlation creep. Pauses strategies when their regime hypothesis no longer holds. Surfaces risk before it surprises you.

Portfolio Risk Sentinel
Sizing, drawdown limits, and alpha-decay defense.

Defends the book day to day. Sizes positions with fractional Kelly, tracks rolling Sharpe, Sortino, and max drawdown across every active strategy, and enforces per-strategy and portfolio-wide risk limits. Auto-pauses any strategy whose live performance decays past threshold so a stale edge can't bleed the book while you're not looking.

Quant Research Lab
Feature discovery and unconventional algos.

Searches for non-obvious features — kernel methods, hidden Markov regimes, factor interactions — and proposes them as research questions. Outputs are validated through the same Monte Carlo + walk-forward gates as everything else, so nothing trades that hasn't earned it.

Extensible

Python SDK & MCP tools

Geniuses run Python in a sandboxed environment with a tg object that wraps the platform — pricebars, time series, indicators, options pricing, portfolio operations, the graph. Every scheduled job has access to the same SDK your team uses interactively.

  • ▸tg.get_prices(), tg.eval_history() for graph price reads, tg.Layer() for scenario overlays, tg.observation_date for historical re-runs
  • ▸tg.register_strategy() to enter the strategy registry and lifecycle
  • ▸tg.manage_portfolio() for paper-first execution and position lifecycle
  • ▸MCP tool surface — Geniuses can call any registered MCP server your team adds (browser, search, custom data feeds, internal APIs)
Discipline

Nothing trades that hasn't earned it

Every registered strategy clears the same gates before it touches paper, and the same gates before paper graduates to live.

  • ▸Monte Carlo simulation — distribution of outcomes, not a single backtest line
  • ▸Walk-forward testing — out-of-sample windows, parameter stability checks
  • ▸Paper-first execution — virtual portfolio with real prices and real reconciliation
  • ▸Alpha decay monitor — rolling performance tracking, auto-pause when the edge stops working
    • Inputs

    Sources & integrations

    Out of the box. Custom feeds slot in via Co-Build.

  • ▸Brokers: IBKR (paper + live), with daily reconciliation
  • ▸Market data: Yahoo Finance EOD + intraday bars
  • ▸Options: chains and SVI vol surfaces (in progress)
  • ▸Custom: any MCP-compatible tool or data source
  • ▸On the roadmap:SEC EDGAR fundamentals, FRED macro series, Kalshi / Polymarket, weather & shipping, inventories, Reddit / RSS sentiment
    • Get in touch